Linkage of National Congenital Heart Disease Audit data to hospital, critical care and mortality national data sets to enable research focused on quality improvement.

OBJECTIVES: To link five national data sets (three registries, two administrative) and create longitudinal healthcare trajectories for patients with congenital heart disease (CHD), describing the quality and the summary statistics of the linked data set.
DESIGN: Bespoke linkage of record-level patient identifiers across five national data sets. Generation of spells of care defined as periods of time-overlapping events across the data sets.
SETTING: National Congenital Heart Disease Audit (NCHDA) procedures in public (National Health Service; NHS) hospitals in England and Wales, paediatric and adult intensive care data sets (Paediatric Intensive Care Audit Network; PICANet and the Case Mix Programme from the Intensive Care National Audit & Research Centre; ICNARC-CMP), administrative hospital episodes (hospital episode statistics; HES inpatient, outpatient, accident and emergency; A&E) and mortality registry data. PARTICIPANTS: Patients with any CHD procedure recorded in NCHDA between April 2000 and March 2017 from public hospitals. PRIMARY AND SECONDARY OUTCOME MEASURES: Primary: number of linked records, number of unique patients and number of generated spells of care. Secondary: quality and completeness of linkage.
RESULTS: There were 143 862 records in NCHDA relating to 96 041 unique patients. We identified 65 797 linked PICANet patient admissions, 4664 linked ICNARC-CMP admissions and over 6 million linked HES episodes of care (1.1M inpatient, 4.7M outpatient). The linked data set had 4 908 153 spells of care after quality checks, with a median (IQR) of 3.4 (1.8-6.3) spells per patient-year. Where linkage was feasible (in terms of year and centre), 95.6% surgical procedure records were linked to a corresponding HES record, 93.9% paediatric (cardiac) surgery procedure records to a corresponding PICANet admission and 76.8% adult surgery procedure records to a corresponding ICNARC-CMP record.
CONCLUSIONS: We successfully linked four national data sets to the core data set of all CHD procedures performed between 2000 and 2017. This will enable a much richer analysis of longitudinal patient journeys and outcomes. We hope that our detailed description of the linkage process will be useful to others looking to link national data sets to address important research priorities. © Author(s) (or their employer(s)) 2022. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ.

We linked five national established, high-quality, data sets using bespoke methods for the preprocessing of identifiers and establishing matches to maximise linkage.

In our final data set, data consistency has been checked at patient level using year and month of birth, postcodes and diagnosis codes, and also clinically sense checked at spell level for spells containing congenital heart procedures.

We created meaningful spells of care for each patient in the data set covering inpatient and outpatient interactions with secondary and tertiary care, covering up to 20 years of life of patients with congenital heart disease (CHD), representing an important step to understanding patient care for people with CHD.

Data completeness, quality and availability were worse in earlier years, meaning that linkage was poorer for earlier eras.

We do not yet have data on hospital care for patients outside England or on longer term adult follow-up for patients whose full CHD history is captured, since most cardiac procedures start in early live—the national CHD audit started on April 2000.

Introduction

Measuring, reporting and learning from patient outcomes should drive quality improvement (QI), but this is particularly challenging for lifelong conditions where outcomes need to be interpreted in the context of different phases of treatment, changing treatment options, changing service provision and the natural evolution of disease.1 2 Given the complex longitudinal care trajectories of such patients, rich data sets and careful multidisciplinary analysis are required to understand how patients interact with health services and to identify relevant outcomes and meaningful variations. These then provide opportunities for more targeted QI. Services for congenital heart disease (CHD) provide one such example. They span a patient’s lifetime, but their quality in the UK is mainly measured by 30-day survival following children’s heart surgery or catheter-based procedures. This is no longer a sufficient proxy and a more sophisticated approach is required.3

Information on patients with CHD, and their utilisation of specialised care services in England and Wales, is not available in a single data set. Since April 2000, the main source of information on the early outcomes of therapeutic paediatric and congenital cardiovascular procedures for patients with CHD in UK has been the National Congenital Heart Disease Audit (NCHDA).4 5 Submission is mandatory for all centres and data quality is subjected to external validation. The key feature of this data set is the detailed recording of cardiac-related diagnosis and procedural information using the European Paediatric and Congenital Cardiac Code short list descriptors.6

By linking NCHDA with other national data sets, both validated registries and administrative, we aimed to build a unique combined data set for understanding patient journeys through the secondary and tertiary healthcare system. The four relevant national data sets are the Paediatric Intensive Care Audit Network (PICANet) for patient admissions to paediatric intensive care units (PICU)7; the case mix programme (CMP) from the Intensive Care National Audit & Research Centre (ICNARC-CMP) for patient admissions to adult intensive care units8; death registrations from the Office for National Statistics (ONS); hospital episode statistics (HES) routine administrative data on admitted patient care (APC), accident and emergency (A&E) attendances and outpatient (OP) appointments at National Health Service (NHS) hospitals in England.9 10

The research project ‘LAUNCHES QI: Linking Audit and National data sets in Congenital Heart Services for Quality Improvement’ aims to: describe patient trajectories through secondary and tertiary care; identify useful metrics for driving QI and informing commissioning and policy; explore variation across services to identify priorities for QI. In this paper, our objective is to describe the methods used to link the NCHDA data to HES, ONS, PICANet and ICNARC-CMP data sets and report the general characteristics, strengths and limitations of the resulting LAUNCHES data set. The process and challenges involved in the application for the approvals needed to link the LAUNCHES data sets have been described elsewhere.11

Methods

Data

The core data set in LAUNCHES is NCHDA,4 5 from which we obtained data for all records between 1 April 2000 and 31 March 2017 (figure 1). Each record relates to a single CHD procedure carried out in public hospitals in England and Wales. Most patients are resident in England and Wales, but patients from Northern Ireland and Scotland and overseas are also represented. NCHDA provides detailed demographic, diagnosis and procedural information for CHD procedures in children and adults as well as short-term survival outcomes (in-hospital and at 30 days).12 Online supplemental table S1 contains all NCHDA fields that we obtained for LAUNCHES.

Figure 1

Data sets and years covered to make up the LAUNCHES data set. Calendar years are displayed at the top of this figure, while the data were obtained by financial years, which run from 1 April to 31 March. A&E, accident and emergency; HES, hospital episode statistics; ICNARC-CMP, Intensive Care National Audit & Research Centre Case Mix Programme; NCHDA, National Congenital Heart Disease Audit; ONS, Office for National Statistics (mortality); PICANet, Paediatric Intensive Care Audit Network.

We applied to link to the following HES data sets (figure 1): APC inpatient (not limited to cardiac) admissions to hospitals in England between financial years 1998/1999 (starting 1 April 1998) and 2017/2018 (ending 31 March 2018); HES OP appointments between financial years 2003/2004 (first year available) and 2017/2018; HES A&E attendances between financial years 2007/2008 (first year available) and 2017/2018.9 10 13 Online supplemental tables S2–S4 contain all HES fields that we obtained for LAUNCHES.

The ONS mortality data are the most complete source for the assessment of patient survival, recording all deaths registered in England and Wales.14 Linked to HES data,15 we obtained the ONS life status of patients of patients resident in England and Wales. See online supplemental table S5 for all ONS fields.

The PICANet contains records for all children admitted to PICU within UK and Ireland.7 We requested all PICANet admissions in England up to March 2017 that could be linked to records in NCHDA (see online supplemental table S6 for all PICANet fields).

The CMP collects data from adult general critical care units in England, Wales and Northern Ireland.8 16 We requested all ICNARC-CMP admissions up to August 2018 that could be linked to records in NCHDA (see online supplemental table S7 for all ICNARC-CMP fields).

The selected HES years correspond to all years of HES data with available HES identifiers (HES IDs) and NHS numbers (see HES Data Dictionary17) at the application time, where HES APC year 1997/1998 was not requested because we were informed that NHS numbers were largely missing (55.5%).

No dates of patient events were requested, other than year and month of birth (online supplemental tables S1–S6). Instead, ages (in years) to 4 decimal places at each event were requested from data providers to facilitate construction of detailed healthcare trajectories (enabling ordering of multiple events on the same day) while minimising identifiability of the linked data.

Data identifiers used for linkage

Table 1 lists the identifiers used for linkage, the data sets each were present in, and any prelinkage processing that was undertaken. NHS numbers have some limitations,18 19 particularly that they are likely to be missing for overseas patients or those from Scotland and Northern Ireland. Hospital identifiers are unique to a patient, and records with the same hospital identifier will relate to the same patient. But hospital identifiers change between hospitals and so are not useful for linking patient records across different hospitals. In the absence of a matching NHS number or hospital patient identifier, we used date of birth, name and postcode to identify records as pertaining to the same patient but only if all three matched across records. We categorised the quality of each identifier for each record as: valid (for linkage), invalid or missing (table 1).

Table 1

Identifiers used for linkage

Identifier	Description and processing undertaken	Data set
NHS number	NHS numbers are 10-digit identifiers assigned to people registered for NHS care in England, Wales, or the Isle of Man. They are assigned to patients soon after birth (since year 2002) or the first time they receive NHS care or treatment.30Processing: removed non-numeric characters and blanks.Invalid values: 10-digit numbers that are all the same; dummy value ‘2333455667’; format ‘n00000000n’ (eg, ‘6000000006’).15Valid values: Not invalid (above) and satisfying the checksum digit check.31	NCHDA, PICANet, ICNARC-CMP, HES/ONS
Hospital patient ID	Hospitals use their own local patient identifiers, which in combination with the centre ID constitute a unique patient identifier that we refer to as ‘hospital patient identifier’. A patient can have multiple hospital identifiers across their records for example, associated with care in different hospitals at different times.Processing: standardised the centre ID values, and removed blanks, leading zeroes and leading/trailing special characters from the local patient identifiers.15Valid values: any value was considered valid.	NCHDA, PICANet
Date of birth (DoB)	Date of birth of the patient is available as recorded in the data setsProcessing: standardised the format to day/month/year (eg, 17/11/2007).Invalid values: Any date after 01/04/2017 or before 01/01/1895. Equal to either 01/01/1901 or 31/12/1899.15Valid values: Not invalid (see above) and a feasible date.	NCHDA, PICANet, ICNARC-CMP, HES/ONS
Name/surname	Processing: converted to upper case; removed prefixes and titles (eg, MISS, MSTR, MASTER, MRS, MS, MR, MAST, DR, SGT, SHEIKHA, SULTANA, SHEIKH, SULTAN), removed generic values (eg, BABY, INFANT, TWIN, TRIPLETS, BOY, GIRL, NAME1, NAME2). Removed special characters (apostrophes and accents).Valid values: non-empty values (after processing the fields).	NCHDA, PICANet
Postcode	Processing: converted to uppercase, removed blanks and special characters (only alphanumeric characters allowed).Valid values: postcodes included in the historical list of postcodes from the Organisation Data Service32 and not corresponding to country postcodes (starting with ‘ZZ’) and not from an NHS trust site.33	NCHDA, PICANet, ICNARC-CMP, HES/ONS

HES, hospital episode statistics; ICNARC-CMP, Intensive Care National Audit & Research Centre Case Mix Programme; NCHDA, National Congenital Heart Disease Audit; ONS, Office for National Statistics (mortality); PICANet, Paediatric Intensive Care Audit Network.

Linkage method

We developed an algorithm to link NCHDA data both internally (to identify records pertaining to the same person within NCHDA) and externally, to records in the other data sets. Our hierarchical method, shown in figure 2, treated NHS number and hospital patient ID as primary identifiers, while date of birth, patient name and postcode were treated as weaker identifiers. The possible linkage states when comparing a processed identifier across two records were:

Figure 2

The linkage algorithm for deciding whether two records pertain to the same patient. A: linkage of NCHDA records internally and to PICANet records. B: linkage of NCHDA to ICNARC-CMP and to HES/ONS. ‘No DoB disagreement’ means that the dates of birth either match (exactly or partially) or one or both of those dates are missing. HES, hospital episode statistics; ICNARC-CMP, Intensive Care National Audit & Research Centre Case Mix Programme; NCHDA, National Congenital Heart Disease Audit; ONS, Office for National Statistics (mortality); PICANet, Paediatric Intensive Care Audit Network.

Exact agreement, if each identifier was valid and they were exactly the same.

Partial agreement only used for valid dates of birth and names and defined in detail below.

Any missing, if either or both identifiers were missing or invalid.

Disagreement, if both values were valid and non-missing but did not match (exactly or partially).

Two valid dates of birth (DoB) were considered to be in partial agreement if either: the two DoB values were no more than 5 days apart; the two DoB values were not the same, but either two components (ie, YYYY, MM or DD) of the two DoB values matched or two components of the two DoB values matched when the MM and DD parts of one of them were swapped. Partial agreement of names occurred between two records if there were previous and current versions of names and at least one matched the other record.

An auxiliary lookup table (online supplemental table S8) between NCHDA organisations and PICUs was used by PICANet when comparing hospital patient identifiers as part of the NCHDA to PICANet linkage (figure 2A), given that the two data sets use different names for centres.

For NCHDA to ICNARC-CMP linkage, two records were matched by ICNARC only if there was exact agreement of NHS numbers and either the DoB did not disagree or postcodes matched exactly (figure 2B). NCHDA to HES/ONS linkage was performed by NHS Digital and required the exact match of NHS numbers (agreement in postcode was reported but not required). See online supplemental table S9 for the HES/ONS linkage method.

Finally, note that all linkages were done at record level. This resulted in many-to-many record matches that were resolved to identify records as pertaining to the same patient across all five data sets once pseudonymised data sets had been received at University College London (UCL).

Data flows

Record-level patient identifiers in the core data set (NCHDA) were sent for linkage via secure transfer to each of the three data controllers for the other four data sets, along with a study-specific pseudonymised record identifier. Each data controller then searched for records within their data sets with matching patient identifiers and returned the pseudonymised, clinical data (without patient identifiers) for all records that had at least one match to an NCHDA record to UCL Clinical Operational Research Unit. We used secure transfer and all data are stored in the UCL data safe haven, which complies with the NHS Information Governance Toolkit. Only pseudonymised study-specific record and patient IDs were shared with or stored at UCL. Linkage results were provided as lists of corresponding pairs of records with a code indicating the quality of linkage for each record-to-record match (concatenated agreement category for each identifier).

Patient-level consistency and quality assurance

The national audit body (National Institute for Cardiovascular Outcomes Research; NICOR) identified unique patients within the NHCDA using the linkage algorithm and then checked for inconsistencies on site as part of data quality assurance. Inconsistencies in DoB (missing values, procedures before birth, different DoB for a same patient) were identified and sent to submitting hospitals for correction and were then revised by NICOR. Cleaned record identifiers were then sent for linkage to the other data processors. An additional internal detailed clinical review was undertaken of pairs of records that were not linked but similar to some extent (eg, those pairs solely agreeing in NHS number) and pairs of records linked but with only moderate agreement in identifiers (eg, pairs with matched names, DoB and postcode but NHS numbers missing) and internal patient categorisation updated.

Both HES and PICANet have their own internal unique patient IDs across records. Pseudonymised versions of these were included in the returned records. We then assessed the level of agreement between the identified patients from the NCHDA and patient identifiers from the linked PICANet and HES data sets. PICANet and HES patients linked to more than one LAUNCHES patient were discussed with each processor and patient categorisation was revised on a case-by-case basis. Numbers of records and patients before, during and after quality assurance will be reported, together with available years of follow-up.

Spells of care and completeness of linkage

Once the linked data set was created, we combined overlapping events into ‘spells of care’. Gaps of less than 24 hours were considered to be overlapping, since times of events were not routinely collected and so records could have a 12-hour uncertainty in either direction. Figure 3 illustrates an example of event records that would be combined into a single (paediatric) spell. Number of spells per year/patient/data set will be reported.

Figure 3

Example of Care spell consisting of several time-overlapping events involving different services. A&E, accident and emergency; HES, hospital episode statistics; ICNARC-CMP, Intensive Care National Audit & Research Centre Case Mix Programme; PICANet, Paediatric Intensive Care Audit Network.

Cardiac surgeries typically require intensive care recovery. Catheter-based interventions and diagnostic procedures are far less likely to require ICU admission. Our first consistency check was to look at how many spells containing a cardiac surgery procedure also contained an accompanying ICU stay, enabling an assessment of the completeness of linkages from NCHDA to PICANet and NCHDA to ICNARC-CMP. While we would not expect 100% of NCHDA surgeries to have a linked record, we would expect a high proportion to. A second consistency check was for HES linkage completeness. We would expect a HES-linked record (either inpatient admission or OP attendance) to be part of the same spell as any NCHDA procedure, as long as the NHCDA record had a valid NHS number. In addition, at least one of the ICD-10 diagnostic codes used within HES for inpatient admissions should denote CHD for HES records linked to NCHDA surgical procedures (a list of valid congenital codes and other cardiac non-congenital codes that are sometimes used for patients with CHD is provided in online supplemental table S10). Summary statistics will be provided on the completeness of linkage per data set and the clinical sense checking of HES linked data.

Patient and public involvement statement

We have patient and public representatives on the independent study advisory group. The advisory group was consulted on linkage design and execution and approved the process.

Results

Quality of identifiers in each data set

The NCHDA data set contained 143 862 CHD records of which 94.7% had valid NHS numbers. Unsurprisingly, the percentage of valid NHS numbers was higher for patients with residence in England (98.8%) or Wales (99.1%) as determined by their postcode at the time of procedure. The breakdown of NHS numbers by residence is given in online supplemental table S11. PICANet records for patients born before 14 October 2001 were available only if they had a PICANet event between 14 October 2014 and 13 October 2019, due to the terms of the PICANet Health Research Authority (HRA) Confidentiality Advisory Group (CAG) approval for processing identifiable information.20 There were 179 791 PICANet records available for linkage, of which 90.5% had valid NHS numbers. Hospital patient identifiers were available for 100% of NCHDA and PICANet records, as were DoB; names/surnames were available for 99.6% and 98.9% of records, respectively, and postcodes were valid for 95.0% and 97.2% of records. ICNARC-CMP had 1 853 568 records of which 88.7% had valid NHS numbers. The total of records and percentage of valid NHS numbers for HES data were: 314 445 082 (93.8%) for HES inpatient, 1 288 711 692 (98.0%) for HES OP and 194 572 279 (93.3%) for HES A&E. We did not know the quality of identifiers in ONS mortality data, which we obtained linked to HES data. The quality of the identifiers improved over time (online supplemental table S12).

Linked data sets before quality assurance

There were 6 408 673 records across the final component data sets before any quality assurance was carried out (online supplemental table S13), with each non-NCHDA record linked to at least one NCHDA record.

Quality of the record-level linkage

The use of a bespoke method for linking NCHDA-NCHDA and NCHDA-PICANet records (figure 2A) allowed us to identify more linked records than had we relied solely on NHS numbers:

95.0% of the NCHDA-NCHDA matches and 92.3% of the NCHDA-PICANet matches were identified by an exact agreement of NHS numbers.

4.9% of the NCHDA-NCHDA and 7.0% of the NCHDA-PICANet matches were identified by exact agreement in hospital patient identifiers (allowing for missing NHS number).

0.1% of the NCHDA-NCHDA and 0.7% of the NCHDA-PICANet matches were identified by other options of our bespoke linkage algorithm.

Patient-level results

There were 47 753 internal NCHDA-linked records (out of a total of 143 862 NCHDA records), representing patients with more than one recorded procedure within the NCHDA data set.

Once patients had been defined across NCHDA records, 649 inconsistencies in DoB affecting 219 patients were detected and corrected. There was a very high level of agreement between the identified patients from the linked PICANet data and the LAUNCHES linkage definition of patients: only seven PICANet patients (0.0% of the 34 507 linked PICANet patients) were linked to two LAUNCHES patients each. Investigation of those cases by each audit resulted in a further minor revision. In a similar exercise, we excluded 88 HES IDs (0.1% of the total 89 098 linked HES IDs) that were linked to two LAUNCHES patient IDs each. It was not possible to determine which HES records corresponded to each patient (mainly because they pertained to twins). Inconsistencies between 42 HES and NCHDA patients linked with disagreement in year–month of birth and postcode were also resolved.

This detailed review of linked NCHDA records resulted in a final total of 96 041 unique patients with a total of 6 381 600 records (table 2). Of those, 66 453 patients (69.2%) had at least one NHCDA record as children (age at procedure under 16), whereas the remaining 29 588 patients (30.8%) had all their NHCDA records as adults.

Table 2

Number of linked records in each data set after quality assurance, by estimated financial year

Financial year	NCHDA	PICANet	ICNARC-CMP	HES inpatient	HES outpatient	HES A&E	Total
1998	0	0	0	16 431	0	0	16 431
1999	0	0	0	19 811	0	0	19 811
2000	6421	15	2	29 113	0	0	35 551
2001	6161	11	1	33 210	0	0	39 383
2002	6137	952	0	36 870	0	0	43 959
2003	7402	3226	0	42 805	132 364	0	185 797
2004	6968	3464	0	45 314	149 544	0	205 290
2005	7684	3828	0	50 097	176 383	0	237 992
2006	8152	4052	6	52 001	195 655	0	259 866
2007	7984	4136	154	56 577	223 402	23 268	315 521
2008	8294	4275	215	59 782	254 476	27 482	354 524
2009	8719	4748	273	65 190	292 972	32 732	404 634
2010	8987	4891	388	69 084	322 196	35 862	441 408
2011	9102	5103	407	70 564	347 096	38 854	471 126
2012	9013	5176	411	70 908	368 160	41 598	495 266
2013	9593	5435	473	71 781	406 805	42 830	536 917
2014	9639	5435	447	72 751	440 554	44 913	573 739
2015	11 492	5546	629	75 959	468 434	47 219	609 279
2016	12 114	5504	686	72 899	476 727	46 885	614 815
2017	0	0	572	51 814	424 473	43 432	520 291
All years	143 862	65 797	4664	1 062 961	4 679 241	425 075	6 381 600

Financial years (running from April to March) were estimated using the ages at events and the estimated date of birth (we took day 15th of the known month of birth as date of birth). The estimation is likely wrong for 27 records from PICANet and ICNARC-CMP with estimated year pre-2002, but we could not fix the needed ages or dates of birth at the time of submission (such inconsistencies are likely to be excluded in future analyses).

A&E, accident and emergency; HES, hospital episode statistics; ICNARC-CMP, Intensive Care National Audit & Research Centre Case Mix Programme; NCHDA, National Congenital Heart Disease Audit; PICANet, Paediatric Intensive Care Audit Network.

A total of 90 678 patients (94.5%) were linked to at least one external data set: 91.5% of patients had some form of HES/ONS record, 35.9% had at least one linked PICANet record and 3.6% had at least one linked ICNARC-CMP record. The main reasons for non-linkage of the remaining 5363 patients (5.6% of all NCHDA patients) were: missing NHS number; residence not recorded or outside England; and/or record from before 2003 when data quality was poorer. The final linked data set covers up to 20 years of life of patients, with a median (IQR) coverage of 12 (6, 16) years for 87 735 patients with no known age of death and 4 (1, 13) years for 8306 patients with known age of death.

Spell-level results

We identified 4 908 153 spells of care for the 96 041 patients in the LAUNCHES data set. Only 2.6% of the spells contained at least one NCHDA procedure compared to the 99.7% of spells that included at least one HES record (799 890 inpatient spells in total). Only 1.0% of spells included at least one PICANet record, and 0.1% of spells included at least one ICNARC-CMP record. Patients had a median (IQR) of 3.4 (1.8, 6.3) spells per year, with a median (IQR) of 0.1 (0.1, 0.3) spells with NCHDA procedures per year. This high level of healthcare interaction was expected in this population, since patients with CHD require regular specialist follow-up.

Sense checking the completeness of the linkage

PICANet

Out of all paediatric cardiac surgeries, 93.9% (42 512/45 265) were linked to an associated PICANet record where linkage was in principle feasible. The corresponding percentage for paediatric catheter-based procedures was 11.2% (2047/18 268).

ICNARC-CMP

Out of all adult cardiac surgeries (resp catheters), 76.8% (906/1180) (resp 2.6%: 69/2610) were linked to ICNARC-CMP when the procedures were post-March 2009 at centres submitting regularly to ICNARC, and where a valid NHS number was recorded. Unfortunately, many hospitals carrying out congenital heart procedures submitted very few records to ICNARC-CMP over the time period of this study. This means that for all cardiac surgeries where ICNARC-CMP data would have been available (post 2009 with a valid NHS number), only 16.5% (1193/7234) were linked to an associated CMP record.

HES/ONS

Out of all NHCDA procedure records (either surgical or catheter) with a valid NHS number and performed in an English public hospital, 95.6% (122 278/127 932) were linked to an associated HES record, mostly inpatient records. ONS age at death was provided for 7228 patients. In a total of 53 769 spells which included both NCHDA surgical procedures and an associated HES inpatient record, 94.6% of HES records had CHD ICD-10 diagnostic codes from online supplemental table S10, 3.8% had only acquired heart diagnoses (plausible miscoding of CHD) and 1.6% had other diagnostic codes.

These consistency checks provide assurance that, where linkage was theoretically possible, we achieved excellent linkage.

Discussion

Principal findings

We have described a bespoke linkage algorithm, alongside quality, completeness and consistency checks, which we used to identify 96 041 unique patients across 143 862 NCHDA cardiac procedure records and to link their records to 65 797 PICU admissions, 4664 adult intensive care admissions and 6 167 277 HES (inpatient, OP and A&E) records.

While most of the linked records were identified using matching NHS numbers, a significant proportion (around 5%) was identified using other identifiers, highlighting the value of using additional identifiers. Close collaboration with each audit and NHS Digital meant that we could further check the quality of the linkage and further refine the identification of unique patients across records, improving the overall quality of the linked data set.

The quality of recorded identifiers used for linkage improved markedly over time as did the quality of resulting linkage. 90 678 (94.5%) patients had records that were linked to at least one other data set. We identified 4 908 153 spells of care for the 96 041 patients. The final linked data set (6 381 600 records) covers up to 20 years of life of patients, with a median (IQR) coverage of 12 (6,16) years for 87 735 patients with no known age of death, and 4 (1, 13) years for 8306 patients with known age of death.

Patients had a median (IQR) of 3.4 (1.8, 6.3) spells of care (either an inpatient stay or an OP event) per year. This frequent interaction with secondary and tertiary care outside of NCHDA procedures (only 2.6% spells of care included an NCHDA procedure) highlights the necessity and value of linking specialised validated procedure-based registry records (NCHDA) to other administrative and audit data sets to understand and potentially improve services for CHD.21 22

Strengths and weaknesses

All linked data sets were national established, high-quality, data sets. We designed a bespoke linkage method and data processors carefully prepared the identifiers for linkage in a consistent way to maximise matching. In our final data set, data consistency has been checked at patient level using year and month of birth, postcodes and diagnosis codes and also clinically sense checked at spell level for spells containing congenital heart procedures.

Each of the data sets used for linkage was available for different years. Additionally, PICANet’s HRA CAG policy of data anonymisation restricted linkage feasibility for some patients, HES data only covered hospitals in England and ICNARC-CMP data set was of limited utility since many specialised adult cardiac intensive care units did not submit to ICNARC-CMP for most or all of the time period. More adult cardiac ICUs submit to ICNARC-CMP every year and so future linkage should be much more complete.

The linked data set covers at most 20 years of life of patients. While this represents an important step to understanding patient care for people with CHD, we do not yet have data on longer term adult follow-up for patients whose full CHD history is captured (ie, those born after 2000), since most cardiac procedures start in early life.

Comparison with other studies

In the UK, the Infant Heart Study linked an NCHDA cohort to PICANet data to explore risk factors for poor outcomes (1 year) after hospital discharge for infants undergoing heart surgery between years 2005 and 2010.23 24 ONS mortality was included as part of NCHDA at that time, and the linkage to PICANet was carried out using just NHS number. A study looking at differences in access to Emergency Paediatric Intensive Care and care during Transport linked together PICANet, ICNARC-CMP and HES/ONS. NHS numbers were the primary identifiers used for matching.25–27 Our bespoke linkage algorithm improved the approach based on NHS numbers, with 7.7% of the total NCHDA-PICANet matches obtained using agreement in other identifiers.

Implications for clinicians and policymakers

The NHCDA database is highly specialised and procedure based. The linked intensive care and hospital data sets provide a much wider and more complete picture of the interactions CHD patients have with secondary and tertiary care throughout their lives. In particular, the OP data means loss to follow-up in transition from child to adult services and/or during adulthood can be explored. The linked data of validated registries with administrative databases will facilitate the identification of appropriate outcomes for reporting and routine monitoring CHD services at all ages, including resource utilisation, and to develop methods of QI that take into account differences in risk across case mix.28

Unanswered questions and future research

The NCHDA data set only contains information for CHD patients that have at least one procedure. This means that when considering overall health service journeys of people living with CHD, we miss those who never have a procedure (either because disease is considered too mild or because it is too severe for correction). The ongoing CHAMPION project will use the National Congenital Anomaly and Rare Disease Registration Service (NCARDRS) data set to estimate the number of children born with CHD or that have an antenatal diagnosis but do not survive pregnancy (termination or in-utero death).28 29 In future, linkage to NCARDRS might allow assessment of outcomes and healthcare journeys for the complete patient cohort.

Conclusion

We successfully linked five national data sets to achieve a large, high-quality combined data set spanning 20 years that will allow rich exploration of the healthcare journeys of patients with CHD. We hope that this detailed description will be useful to others looking to link national data sets to address important research priorities. While challenging, researchers, data controllers and data processors should continue to encourage and facilitate data linkage to enable generation of valuable new knowledge and insights.